EVIDENCE TRAIL
AI-source disclosure UI — visible AI labelling at the point of action
Verbatim excerpts from the upstream sources cited on the mitigation page, with what each source does and does not prove. The strongest legal mandate is EU AI Act Article 50 (paragraphs 1 and 2), which separately covers interaction-level disclosure and machine-readable content marking. NIST AI 600-1 MS-2.7-003 is the closest action-level analogue from the US federal framework.
Last cross-checked against upstream sources: · 8 sources
References
Each entry shows what the source supports and what it does not prove.
EU AI Act — Article 50, paragraph 1
Article 50 — Transparency obligations for providers and deployers of certain AI systems — paragraph 1
"Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use."
Supports: Verbatim legal mandate for interaction-level AI disclosure (the third labelling layer in this control). Applies to any AI system that interacts directly with natural persons, scoped to EU jurisdiction.
Does not prove: Does not specify the visual format of the disclosure, its persistence, or how friction should scale with action irreversibility. The exemption for "obvious" AI use is fact-specific and not self-executing.
EU AI Act — Article 50, paragraph 2
Article 50 — Transparency obligations for providers and deployers of certain AI systems — paragraph 2
"Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated."
Supports: Verbatim mandate for content-level machine-readable marking of AI-generated outputs — the technical foundation for the content-level labelling layer of this control.
Does not prove: Paragraph 2 addresses machine-readable marks (watermarks, C2PA manifest), not the human-visible UI badge. Human-visible labelling of deepfakes is covered separately in paragraph 4. Exceptions apply for assistive editing and non-substantial alterations.
NIST AI 600-1 — Generative AI Profile (GOVERN 1.2)
GOVERN 1.2 — Action GV-1.2-001
"Establish transparency policies and processes for documenting the origin and history of training data and generated data for GAI applications to advance digital content transparency, while balancing the proprietary nature of training approaches."
Supports: Names "digital content transparency" as a governance policy requirement for GAI applications, providing an upstream policy anchor for content-level labelling practices.
Does not prove: GV-1.2-001 addresses data-origin documentation for governance purposes, not the user-visible label at the point of action. Helmwart operationalises this upstream policy requirement as a concrete UI control.
NIST AI 600-1 — Generative AI Profile (MEASURE 2.7)
MEASURE 2.7 — Action MS-2.7-003
"Conduct user surveys to gather user satisfaction with the AI-generated content and user perceptions of content authenticity. Analyze user feedback to identify concerns and/or current literacy levels related to content provenance and understanding of labels on content."
Supports: Explicitly references "understanding of labels on content" as a measurable outcome in an AI security and resilience context. The closest NIST 600-1 action to directly naming user-facing AI labelling.
Does not prove: MS-2.7-003 is a measurement action (survey users), not a mandate to display a label. It presupposes labels exist and asks organizations to measure whether users understand them. Does not specify label format or friction levels.
Coalition for Content Provenance and Authenticity (C2PA) — Content Credentials Standard
C2PA mission statement / content credentials overview
"Content Credentials work like a nutrition label for digital content, giving a peek at the content's history available for anyone to access, at any time."
Supports: The "nutrition label" framing directly maps to the content-level labelling layer of this control. C2PA is the leading open standard for machine-readable AI-content provenance marks required by EU AI Act Art. 50(2).
Does not prove: C2PA addresses machine-readable provenance metadata — it is a technical substrate for labels, not a UI pattern prescription. Whether and how the C2PA manifest surfaces as a human-visible badge is an implementation decision above the standard.
NIST AI 100-1 — AI Risk Management Framework 1.0
No verbatim excerpt pulled yet — open the original to verify the cited section.
Supports: MAP-5.2 establishes the practice of regular engagement with AI actors to integrate feedback about impacts — the feedback loop (user-confirmation latency, override rate) that this control uses to calibrate label friction over time.
Does not prove: MAP-5.2 as published covers stakeholder feedback integration, not user-facing AI source labelling per se. The MDX cites MAP-5.2 as a user-facing labelling reference; that is an overreach — MAP-5.2 supports the feedback-calibration rationale, not the disclosure mandate. No verbatim excerpt pulled for the labelling claim.
MITRE ATLAS — AML.M0021 Generative AI Guidelines
No verbatim excerpt pulled yet — open the original to verify the cited section.
Supports: Names policy-level safety controls — including transparency constraints baked into system instructions — as a recognized mitigation class for generative AI systems. AI-source disclosure UI is one operationalisation of such a policy constraint.
Does not prove: AML.M0021 covers system-prompt and policy-level constraints broadly; it does not prescribe user-visible disclosure labels or action-level friction mechanisms specifically. No verbatim description text was retrievable from the ATLAS catalogue at time of cross-check.
MITRE ATLAS — AML.M0034 Deepfake Detection
No verbatim excerpt pulled yet — open the original to verify the cited section.
Supports: Names synthetic-content detection as a mitigation for adversarial AI content — the defensive complement to this control. Where this control labels AI-generated content proactively, AML.M0034 addresses post-hoc detection when provenance has not been disclosed.
Does not prove: AML.M0034 is a detection-side control, not a UI disclosure control. It does not prescribe interaction-level or action-level labelling. No verbatim description text was retrievable from the ATLAS catalogue at time of cross-check.